Address unintended jump of calculated cwnd when leaving slowstart

[rscheff]

Summary

D23655 addresses jumps of cubic cwnd due to outdated t_last_cong
data, but initializing K to zero introduced a different miscalculation,
where cwnd would suddenly jump to the recorded max_cwnd instead of
gradually increasing either using the tcp-friendly equation, or the
concave growth region.

Properly calculating K instead of resetting it to zero resolves this issue.

[rscheff]

• calc ticks_since_cong properly

[rscheff]

• record going out from slowstart at the earliest possible moment

[cc]

Tested with my results: https://github.com/ccui1000/freebsd-tcp/wiki/test-D25746

Other than the comment to complement, I have no issues about this change.

sys/netinet/cc/cc_cubic.c

140

Suggest complement the comment above this added condition.

[rscheff]

• update comment to correctly describe conditionals

[cc]

Looks good. Thanks for picking this up.